AI/LLM Speech & Audio AI Evaluation Specialist (Int. Voice)

MG · New Delhi, Delhi, India · Full Time · Onsite

Posted

About The Opportunity A fast-growing AI innovation lab focused on next-gen voice and speech intelligence, we build and evaluate cutting-edge Large Language Models (LLMs) for real-world conversational AI applications. Our work spans multilingual voice agents, prosody modeling, and context-aware synthesis—serving global clients across fintech, healthcare, and telco sectors. We move fast, validate rigorously, and ship voice experiences that sound human, behave intelligently, and scale reliably. Role & Responsibilities Evaluate and benchmark LLM-driven speech outputs across fluency, naturalness, emotional tone, and contextual accuracy using standardized metrics and custom rubrics.Design and run A/B tests for voice models across accents, registers, and cultural contexts to improve perceived quality and user trust.Annotate, transcribe, and label voice datasets for training and evaluation—focusing on turn-taking, intent alignment, and paralinguistic features.Collaborate with ML engineers to translate evaluation insights into fine-tuning signals and data augmentation strategies.Document voice quality trends, failure modes, and improvement pathways in structured reports for cross-functional review.Stay current with research in speech AI, prosody, and LLM evaluation—applying new methods like pairwise preference ranking or automated metrics (e.g., MOS, WER, CER). Skills & Qualifications Must-Have Python Speech-to-Text Text-to-Speech Speech Evaluation Metrics LLM Evaluation Audio Annotation Tools Conversational AI Multilingual Speech Preferred Prosody Analysis Paralinguistic Features Voice UX Design Benefits & Culture Highlights Work on bleeding-edge voice AI projects with global impact.Collaborative, research-driven culture with access to top-tier LLM tooling.On-site innovation lab environment in India with flexible hours and peer learning. Skills: multilingual,text-to-speech,research,prosody,edge,python,speech,models,metrics

Apply on MG's site